The model is trained over the lookback window, and tested out-of-sample on future data.
The length of the lookback window determines how quickly the model adapts to new information.
Backtesting allows determining the optimal length of the lookback window.




